CN1122256C - Method and device for coding audio signal by 'forward' and 'backward' LPC analysis - Google Patents

Abstract

The invention concerns a method and a device for coding an audio signal by 'forward' and 'backward' LPC analysis, the coding being produced by 'forward' LPC filtering for non-stationary zones and on a synthetic signal based on 'backward' LPC filtering for stationary zones. For each current LPC block (Bn) (10), the degree of stationarity of the digital audio signal is determined (11), and a 'forward' or 'backward' LPC analysis selection value is established (12) based on a decision function, according to the stationarity parameter. A 'forward' or 'backward' LPC analysis criterion (13) is applied to the analysis selection value to produce the audio signal coding before proceeding to the next LPC block. The invention is applicable to mobile radiotelephony, to the production, memorisation of audio recordings, to the transmission by satellite and wideband telephony.

Description

Method and apparatus with forward direction and reverse linear predictive coding analysis of encoding sound signal

The present invention includes usefulness " forward direction " and " oppositely " lpc analysis process and device to sound signal (such as speech signal) coding.

At present, consider very leap ahead of user's issued transaction, be for the target of the coding techniques of sound signal (especially speech signal), reduce transmission output condition in, transmit these signals with digital form, with special assurance to being suitable for transmitting the Network Management of these signals.

In used coding techniques, be called as coding techniques that LPC (linear predictive coding) analyzes and comprise the sound signal that will encode is carried out linear prediction, the way of using continuous piece to this signal to apply the linear filtering prediction is carried out coding temporarily.

In above-mentioned technology, the technology that is called as CELP (code excites linear prediction) coding is the most general technology and some best performances is provided.Other technologies (such as being called as MP-LPC (Multi-pulse LPC) or VSELP (vector with excite linear prediction) technology) are similar to the CELP coding.

Above-mentioned coding techniques is called " analyzing with integrated approach ".Especially, they can make the transmission of audio signals output that belongs to the telephone frequency bandwidth reduce to 16kb/s from 64kb/s (MIC coding) under the help of CELP coding techniques, and after adopting the later development of this coding techniques even can reduce to 8kb/s, after transmission and decoding, aspect the quality of the sound of rebuilding without any appreciable reduction.

Especially, the field of these coding techniquess particular importance is the field of mobile phone.In this application, to the restriction of frequency span and the increasing extremely fast of number of users that each mobile phone user allows, make and coding output must be done corresponding the reduction that meanwhile, the requirement of user session sound quality continues to increase.Other applications of these coding techniquess for example relate to numerical data storage (these data represented these signals on memory carrier), be used for high-quality phone, multimedia via satellite or digital transmission that video or audio conferencing are used.

The linear prediction filter that uses in above-mentioned technology is to be called under the help of analysis module of " lpc analysis " to obtain, and described " lpc analysis " operated continuous digital signal piece.According to the rank preface of analyzing (that is, according to the number of filter factor), these wave filters can be more or less simulated the contour shape of the frequency spectrum of wanting encoded signals reliably.Under the situation of speech signal, these frequency spectrum profiles shapes are called formant (tormant).

Yet for the required the second best in quality coding of most of current application, so the wave filter of determining is not enough to signal is done desirable simulation.Therefore, the residual signal of linear prediction is encoded be absolutely necessary.A kind of mode of operation relevant with the linear prediction residual signal is by the special employing of the preceding LD-CELP that has mentioned (low delay CELP) coding techniques institute.In the case, simulate residual signal by the waveform of taking from code page at random, and it be multiply by yield value.For example, the MP-LPC coding techniques is simulated this residual signal under the help by the variable position pulse that changes through separately yield value, and the VSELP coding techniques adopts the linear combination of the pulse vector of taking from suitable form to realize this simulation.

To provide the illustrative summary description of the method for operating of lpc analysis (particularly " oppositely " LDC analyzes and " forward direction " lpc analysis) below.

Total envelope of frequency spectrum is simulated with the short-term synthesis filter that constitutes the LPC wave filter, and its coefficient is simulated with the linear prediction of the speech signal that will encode.This LPC wave filter (it is an autoregressive filter) has the transport function of formula (1) form $A\left(z\right)=1-\underset{i=1}{\overset{p}{\mathrm{\Σ}}}{a}_i{z}^{-i}$

Here ρ points out filter coefficient α _iTitle and the rank of the linear prediction of application, z points out the variable z through conversion of frequency space.

The estimation factor alpha _iA kind of method be included in the criterion of using in the scope of analysis length of speech signal the error prediction signal energy of this speech signal that minimizes.

In fact, the analysis length by the digital language signal of continuous composition of sample is N the sample that constitutes a coded frame.So the energy of error prediction signal is determined by formula (2): $\mathrm{E\ρ}=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}[s\left(n\right)-\underset{i=1}{\overset{P}{\mathrm{\Σ}}}{\mathrm{\α}}_i(n-1){]}^2---\left(2\right)$

The capable sample of n in the frame of N sample of S (n) expression here.

In the block-by-block cataloged procedure, can advantageously be divided into coded frame the LPC piece of subframe or adjacency.So analysis length N surpasses every length, in order that can consider the sample in a certain number of past, or (if available) sample in the future, its way and the cost of being spent are to postpone suitable coding.

When the PLC analytic process is when the piece of the present frame of the speech signal that will encode is carried out, analysis is called " forward direction " LPC, and its coding carries out in encoder level " in real time ", promptly, carry out during the piece of present frame, its unique processing delay is to be introduced by the calculating of filter coefficient.This is sharp analyzes and comprises the calculated value of filter coefficient is sent to code translator.

Is the target of standard UIT-T G728 with 16kb/s at " oppositely " lpc analysis that LD-CELP uses.This analytical technology is included is not that the piece of the present frame of the speech signal that will encode is made lpc analysis, but integrated signal is carried out lpc analysis.This integrated signal should be understood that this lpc analysis is actually that integrated signal to the piece before the current block carries out, because can obtain with encoder level and code translator level simultaneously.This in scrambler and code translator in operation can avoid the value of the LPC filter coefficient that obtains in scrambler is transferred to code translator from scrambler.For this reason, " oppositely " lpc analysis can be exempted transmission output, and the output of so exempting for example can be used for enriching excitation codepage in CELP coding situation." oppositely " lpc analysis also allows to increase the rank of analysis; In the situation of LD-CELP scrambler, the LPC filter coefficient can reach 50, and by comparison, great majority use the scrambler of " forward " lpc analysis to have only 10 coefficients.

So the proper operation of " oppositely " lpc analysis needs following condition:

-the second best in quality integrated signal, it is in close proximity to the speech signal that will encode, considers the quality of current C ELP scrambler, and it comprises the sufficiently high coding output of speed (being higher than 13kb/s).

-less frame and block length, this is because at the signal of having analyzed with want to have between the encoded signals delay of a piece.Therefore, compare with the stationary time of the speech signal that will encode should be shorter for the length of frame and piece;

-by few transmission error of introducing, transmission reliability between scrambler and code translator and maintenance are transmitted the integrality of data.One when integrated signal and the speech signal that will encode have marked difference, and scrambler and code translator stop to calculate same wave filter, thereby bigger not conforming to may occur, can not get back in scrambler and code translator, calculate the significant similarity of wave filter.

Because the relative merits separately of the lpc analysis of above-mentioned " oppositely " and " forward direction " type, S.PROUST, C.LAMBLIN and D.MASSALOUX have delivered the paper that is entitled as " the low CELP of the delay coding of dual rate (8kbits/s/16kbits/s) that uses the reverse/forward direction self-adaptation LPC prediction that mixes " and (have seen Proc.IEEEWorkshop Speech Co.Telecomm., Sept.1995, pp 37-38), proposed a kind of technology in this article, this technology includes selectively contact " oppositely " and " forward direction " lpc analysis.

The above-mentioned condition of considering the correct function of " oppositely " lpc analysis illustrates, and when when significantly being lower than 16kb/s and transmitting output function, singly has such analysis to present described restriction.Except the quality of integrated signal descends (this has reduced the performance of LDC wave filter), in order to reduce transmission output, usually needing the order of magnitude is 10 to 30ms bigger LPC frame length operation.Therefore as can be seen, under these conditions, when frequency spectrum transition and, more generally, not regional so stably, especially performance depreciation can take place, because for general signal (such as music signal) very stably, " oppositely " lpc analysis is analyzed than " forward direction " PLC sizable advantage.

The purpose of getting in touch above-mentioned two types lpc analysis is to reduce these shortcomings and be increased in advantage intrinsic in every kind of analysis:

-" forward direction " lpc analysis is used for the coding to transition and non-stationary zone;

-" oppositely ", lpc analysis was used for the analysis to plateau region basically.

In addition; the LPC frame that the LPC frame of " forward direction " lpc analysis coding is introduced by " oppositely " analysis of encoding allows scrambler to restrain to identical integrated signal again under the situation of loading error occurring with code translator, therefore provides the much bigger error protection that goes out than single with " oppositely " lpc analysis coding.

Generally, " forward direction " of above-mentioned mixing-" oppositely " lpc analysis comprises carries out two lpc analysis, and one is the speech signal that will encode or " forward direction " lpc analysis of audio frequency, and another is to integrated signal " oppositely " lpc analysis.

For each LPC piece, calculate two wave filters, respectively these filter table are shown " forward direction " LPC wave filter and " oppositely " LPC wave filter.So according to whether steadily the make one's options process of the wave filter that is applied to the LPC piece of signal.Two different criterions of this process need:

-according to first criterion of the prediction gain of wave filter;

-calculate continuously " forward direction " LPC wave filter between second criterion of distance parameter.

For each criterion of these two criterions, set up threshold value.First criterion:

If the distance between the prediction gain of " oppositely " and " forward direction " LPC wave filter greater than first threshold, is then selected " oppositely " LPC wave filter.

Second criterion:

If to the vector of the parameter of representing two approaching " forward direction " LPC wave filters calculate distance be lower than second threshold value, then for the present analysis of doing with " oppositely " lpc analysis pattern, forbid switching from " oppositely " lpc analysis pattern to " forward direction " lpc analysis pattern, too little distance characterizes a plateau region more or less, owing to this reason, it is suitable avoiding changing the lpc analysis pattern.Calculate distance be Euclidean distance between the spectral line of the language that will encode or sound signal.

Can find more detailed description in the above in the paper of mentioning of delivering by S.PROUST, C.LAMBLLIN and D.MASSALOUX about the lpc analysis method of above-mentioned mixing.

Analysis operation method to mixing above-mentioned is done deep research, can draw following important shortcoming:

-for some signal, the prediction gain value of " forward direction " and " oppositely " LPC wave filter may be in the vibration up and down of first threshold.This phenomenon causes the unexpected and frequent change from " oppositely " LPC wave filter to " forward direction " LPC wave filter, perhaps change conversely.The filtering of Yin Ruing is discontinuous thus becomes the reason that integrated signal has quite big deterioration, and does not interrelate with the real frequency spectrum change that the language that will encode or audio frequency are believed in most of times;

-when coding output was low, whether the optimum value of the first threshold that should set up according to wanting encoded signals steadily and significantly to change, when coding output is hanged down then more very.For coding delay, perhaps when transmission output drop, clearly difference is arranged between the coding mode of music signal and speech signal corresponding to 10 to 30ms LPC frame; Main " forward direction " lpc analysis that uses.

Because music signal is very stably, so even can use " oppositely " lpc analysis for long LPC frame.Yet in the situation of speech signal, high zone stably has only the very short duration, thereby the stroke of their " oppositely " lpc analysis pattern is of short duration, so cause the undesired filtering transition that makes the coding quality reduction.So scrambler no longer can be proofreaied and correct the phenomenon of being switched the uncontinuity generation of introducing by wave filter.

-provide best subjective quality thereby the LPC wave filter that the frequency spectrum of wanting encoded signals is done best simulation is not usually always had the wave filter of optimum prediction gain.Therefore, some switching from a lpc analysis pattern to another analytical model that interrelates with instantaneous judgement is useless.

The objective of the invention is to solve above-mentioned shortcoming, its way is used process and device for digital audio encoding by special " forward direction " and " oppositely " lpc analysis.

Another object of the present invention also is to use a process, and it dynamically adapts in " forward direction " the function that makes one's options between lpc analysis and " oppositely " lpc analysis according to the steady degree of wanting encoded signals.

Another purpose of the present invention also is to use a process, it is according in the height discriminating between signal (such as music or figure viewed from behind noise) and other signals (such as language) stably, dynamically adapt to above-mentioned selection function, to allow making only encoding process by " oppositely " lpc analysis and " forward direction " lpc analysis respectively.

A further object of the present invention is, one ought make above-mentioned only coding selects, for given type or have given feature want encoded signals, prevent any unexpected switching to non-selected lpc analysis pattern, therefore prevent transition from " forward direction " LPC wave filter to " oppositely " LPC wave filter, and opposite transition, this transition has the tendency that reduces the integrated signal quality of reproducing.

The dynamic self-adapting process that also has a purpose to be to use above-mentioned selection function of the present invention, the change that makes the lpc analysis pattern by this is reliably corresponding to the change of the stationarity of wanting encoded signals, and the chance that interrelates with the simple cross effect of first and second threshold values is very low thus.

Method and apparatus (they are purposes of the present invention) for digital audio signal coding uses double analysis respectively according to the criterion of selecting between " forward direction " and " oppositely " lpc analysis, with the encoded signal that generation is transmitted, this signal comprises LPC filtering parameter and the coding residual signal of analyzing determination information and not transmitting.Digital audio and video signals is subdivided into frame (sequence with piece of the sample that ascertains the number) again, and uses " forward direction " LPC wave filter and integrated signal to realize respectively for the non-stationary zone the coding of this digital audio and video signals.

They are noticeable, because they comprise respectively and allow:

-determine the stationarity degree of digital audio and video signals according to stationarity parameter (its value is between maximum stationary value and minimum stationary value);

-according to the analysis selective value of stationarity parameter foundation based on decision function;

In order to the non-stationary regional code of digital audio and video signals with " oppositely " LPC filtering the plateau region of integrated signal to be encoded, be applied to LPC filtering analyzing selective value with " forward direction " LPC filtering.

This method of operating can preferentially remain on " forward direction " or " oppositely " LPC filter patterns according to the stationarity degree of digital audio and video signals, and the number that can limit the switching from a kind of filter patterns to another kind of filter patterns or switch conversely.

Method and apparatus as the object of the invention not only can be applicable to field of mobile telephony, and can be applicable to the high-quality phone aspect that generation and rerecorded disc, satellite transmits and multimedia video or audio conferencing are used.

Below by reading instructions and checking that design can be so that understand the present invention, wherein:

-Fig. 1 illustrates the illustration figure of step with the form of general flow chart, and these steps allow to finish coding (it is a purpose of the present invention);

-Fig. 2 a illustrates the total process flow diagram that calculates the step of stationarity parameter for each current LPC piece;

-Fig. 2 b illustrates a kind of particularly advantageous method, in order to carry out the key step according to the calculating stationarity parameter of Fig. 2 a;

-Fig. 2 c illustrates the details of execution graph 2b, and illustrate in greater detail for obtain stationarity parameter regulate in the middle of the details of step of value of stationarity parameter;

-Fig. 2 d and 2e illustrate first and second example that adjustment function is used respectively, and their allow the regulated value according to stationarity function in the middle of the fiducial value calculating of " forward direction " and " oppositely " LPC filter gain;

-Fig. 2 f illustrates the process flow diagram of some steps as an example, it can use decision function and or " oppositely " lpc analysis selective value;

-Fig. 3 goes out out the sketch plan of scrambler with the form of function square frame, and this scrambler can be according to purpose of the present invention to audio-frequency signal coding;

-Fig. 4 illustrates the sketch plan of code translator with the form of function square frame, and this code translator can be to adopting the sound signal decoding of scrambler shown in Figure 3.

Be given in the more detailed description of digital audio signal coding below in conjunction with Fig. 1, described method has been used the double analysis (this is a purpose of the present invention) to the coded signal that transmits respectively according to the criterion of selecting between " forward direction " and " oppositely " lpc analysis.

Put it briefly, it illustrates, and the encoded signal portion that the is transmitted ground of writing s_cn (t) comprises the LPC filtering parameter, and the lpc analysis decision signal.In addition, the coding remaining information resn (t) that does not transmit is useful for finishing cataloged procedure.

Digital audio and video signals is divided into LPC frame (they are LPC pieces that are), for convenience of description for the purpose of, each piece is designated as Bn, and definite sample number N is arranged.

An aspect of cataloged procedure (it is a purpose of the present invention) comprises carries out aforesaid coding for digital audio and video signals, wherein, integrated signal regional and that obtain from the coding residual signal uses " forward direction " LPC filtering for non-stationary, and uses " oppositely " LPC filtering for plateau region.

A noticeable especially aspect of this method (it is a purpose of the present invention) comprises, for each current block for a series of current blocks that form present frame, set up " forward direction " or " oppositely " LPC filtering and select criterion, as shown in Figure 1, each current block (being designated as Bn) can obtain in initial step 10, with in step 11,, determine that numeral is the stationarity degree of sound signal according to steady parameter (being designated as STAT (n)).This stationarity parameter presents a digital value, and it (is designated as STAT in maximum stationarity value _M) and minimum stationarity value (be designated as STAT _m) between.

By convention, and do not damage the generality of cataloged procedure (it is a purpose of the present invention), for signal very stably, stationarity parameter presents maximal value STAT _M, and for extremely jiggly signal, stationarity parameter presents minimum value STAT _m

After above-mentioned steps 11, coding method (it is a purpose of the present invention) comprises step 12 use stationarity parameter STAT (n), sets up the lpc analysis selective value.Logically, this analyzes selective value corresponding to selection of " forward direction " lpc analysis or the selection of " oppositely " lpc analysis.Be designated as d analyzing selective value _n(n), and from special decision function (being designated as Dn) draw.

After testing procedure 13 is connected on above-mentioned steps 12, step 13 allows the LPC filtering application is analyzed selective value dn (n), it is represented by C, to carry out coding for digital audio and video signals, its way is to do " forward direction " LPC filtering for the non-stationary zone of digital audio and video signals, and does " oppositely " LPC filtering for the plateau region of integrated signal.

The execution of decision function Dn and above-mentioned analysis selective value dn (n) has formed the particularly advantageous aspect of the cataloged procedure of purpose of the present invention, because can be according to the stationarity degree of sound signal, preferentially remain on a kind of LPC filter patterns (" forward direction " or " oppositely "), and limit the number that switches to another filter patterns from a kind of filter patterns, or the number that switches conversely.

Put it briefly, the decision function of carrying out and be designated as Dn in step 12 is an auto-adaptive function, and it upgrades according to stationarity parameter for each current block.

Can preferentially remain on a kind of LPC filter patterns (" forward direction " or " oppositely ") according to the stationarity degree of digital audio and video signals to the renewal of auto-adaptive function is feasible, thereby limited from a kind of filter patterns and changed number, or the number that switches conversely to another kind of pattern.

Say in particular, the analysis selective value dn (n) that sets up according to above-mentioned decision function Dn is corresponding to the LPC filter patterns preferred value of (" forward direction " or " oppositely "), and corresponding to another preferred value; This value is in fact represented the value that lacks the priority that is back to " oppositely " or " forward " LPC filter patterns.

Preferred value as the LPC filter patterns, analyzing selective value dn (n) for example can be corresponding to a logical value, for example, the true value of this logical value (value 1) is corresponding to selection " oppositely " LPC filtering, and the benefit value of this true value (value 0) is corresponding to selecting " forward direction " LPC filtering.So as can be seen, trial function in the step 13 is summarised as the test value of the logical value of above-mentioned analysis selective value, to guarantee that in step 14 " oppositely " LPC filtering is used for wanting the plateau region of encoded signals, perhaps " forward direction " LPC filtering is used for the non-stationary zone in step 15, follow 14a and 15a in steps in above-mentioned steps 14 and 15 back, they are back to next piece, for n=n+1, are designated as Bn+1.

Though be expressed as logical value analyzing selective value dn (n), should be understood that this logical value can be associated with preferred value, and the probability of filter patterns set up especially by decision function Dn.Can find out especially, for each current block Bn, can be corresponding for this probable value of probable value (its scope is between 0 and 1) of " oppositely " LPC filtering with true logical value, and logical value 0 can corresponding to above-mentioned between 0 and 1 the benefit value of probable range.The many continuous filtering that this probability depends in identical filter patterns is judged.

The working method of decision function Dn can in fact be associated filter patterns priority with logical variable dn (n), and for each current block Bn, in time and self-adaptation.

Put it briefly, make the adaptive purpose of decision function Dn be, in order to avoid unnecessary switching as far as possible from a kind of filter patterns to another kind of filter patterns, consider total steady degree of wanting encoded signals, preferentially to choose " oppositely " LPC filter patterns gradually, perhaps, on the contrary, choose " forward " LPC filter patterns, see which kind of pattern work gets better.

Say in particular:

-encoded signals steady more, then preferential more " oppositely " lpc analysis of choosing of decision function Dn limits as far as possible and switches to " forward direction " LPC filter patterns;

-on the contrary, encoded signals more not steady, then preferential more " forward direction " lpc analysis of choosing of decision function is limited to any switching of " oppositely " LPC filter patterns as far as possible.

The more detailed description of carrying out specific decision function will be provided in the aft section of instructions, this decision function can be adapted to according to the value of stationarity parameter STAT (n).

Now provide and the preferential Calculation Method of describing the stationarity parameter STAT (n) relevant with each current LPC piece Bn in conjunction with Fig. 2 a.

According to above-mentioned accompanying drawing, the step 11 that comprises the stationarity degree of each the current block Bn that determines digital audio and video signals comprises: from the initial value arbitrarily of stationarity parameter, as as shown in the step 110 of Fig. 2 a, this arbitrary value is designated as STAT (0), in step 111, current block Bn hereto, function as the stationarity parameter value of the continuous analysis selective value of determining number and the piece before current block, stationarity parameter value in the middle of calculating, this value is designated as STAT* (n), and these lpc analysis selective values are designated as d _N-1(n-1) ..., d _N-p(n-p), they are that different continuous blocks prior to current block Bn for the sequence of LPC piece obtain, and the stationarity parameter value of the piece before current block is designated as STAT (n-1).In the step 111 that is shown in Fig. 2 a, (be designated as d with these previous values _N-1(n-1) to d _N-p(n-p)) be correlated with, provide the function of the previous analysis selective value that ascertains the number.The initial arbitrary value STAT (0) of stationarity parameter for example can with above-mentioned stationarity parameter maximal value and the minimum value (STAT in the instructions _MSTAT _m) mean value equate.

Follow step 112 in above-mentioned steps 111 back, this step comprises according to the value of the prediction gain of " forward direction " and " oppositely " wave filter or the analytical model of the frame before present frame, stationarity parameter value in the middle of regulating.In the step 112 of Fig. 2 a, above-mentioned function is designated as g, and (Gpb), here for the frame before present frame, Gpf is the prediction gain of " forward direction " LPC wave filter for STAT* (n), Gpf, and Gpb is the prediction gain of " oppositely " LPC wave filter.In step 112 (that is, comprising the step of stationarity parameter value in the middle of regulating), the stationarity parameter value STAT (n) of current LPC piece Bn is provided by the value of formula (3)

(Gpb), it is corresponding to the middle stationarity parameter value through regulating for STAT* (n), Gpf for STAT (n)=g.

The calculation procedure 111 and the more detailed description that comprises the step 112 of regulating this parameter value of stationarity parameter STAT* (n) in the middle of providing in conjunction with Fig. 2 b now.

According to above-mentioned accompanying drawing,, in this step, obtain the stationarity parameter value STAT (n-1) relevant and analyze selective value d with the piece Bn-1 before current block Bn from initialization step 1110 _N-1(n-1), step 111 is included in the step 1111 carries out a step, and it comprises at current block B _nPiece B before _N-1, differentiate that it is that " forward direction " still is " oppositely " lpc analysis pattern.Shown in Fig. 2 b, this discriminating step 1111 can comprise for analyzing selective value d _N-1(n-1) testing procedure sees whether it is relevant with value of symbol " fwd " or logical value 0, and the latter is corresponding to the benefit value of true logical value.

For the negative response of above-mentioned test, that is, and for any B before current block Bn with " oppositely " lpc analysis pattern analysis _N-1, the step of stationarity parameter comprises the previous frame number of determining in succession with " oppositely " lpc analysis pattern analysis (being designated as N_BWD) in the middle of the calculating in step 1113; In step 1114, it comprises does odds ratio to a previous frame number and an arbitrary reference value (being designated as Na) then, and latter representative is in succession with the frame number of " oppositely " LPC pattern analysis.

For the odds ratio of doing at testing procedure 1114, if sure response is arranged, then calculation procedure is included in step 11114b and adds a determined value to the value STAT (n-1) that middle stationarity parameter value STAT* (n) composes with the stationarity parameter of the piece before current block, this value depends on first arbitrary value of the frame number that expression analyzes in succession (that is, in succession with the previous frame number N_BWD of " oppositely " lpc analysis pattern analysis).At step 1114b, depend on that the determined value of first arbitrary value is designated as f _n(N_BWD).During above-mentioned steps, as can be seen, the middle stationarity parameter value STAT* (n) of current LPC piece Bn is so ratio and previous piece B _N-1Identical stationarity parameter is worth increase accordingly.

If negative response is arranged, then at step 1114a, to current block B for the odds ratio in compare test step 1114 _nThe value STAT of the stationarity parameter of piece before (n-1) composes with middle stationarity parameter value STAT* (n).

Yet, for each previous piece B with the pattern analysis of " forward direction " lpc analysis _N-1, for the positive response of testing procedure 1111, be used to calculate the step 111 of intermediate stable parameter, shown in Fig. 2 b, be included in step 1112 and determine, by the current block B of (n-2) row according to test criteria _N-1Whether the piece before the piece before occurs promptly, whether having lpc analysis selective value d from the transition of " oppositely " lpc analysis pattern to " forward direction " lpc analysis pattern _N-2(n-2)=value of symbol " bwd ", (as mentioned above, its logical value is 0).Point out to exist such at current block B for the positive response of testing procedure 1112 _N-1Piece before the piece before is from the transition of " oppositely " analytical model, and points out there is not such transition for the negative response of above-mentioned testing procedure 1112.

For the positive response of above-mentioned testing procedure 1112,, calculation procedure 111 use the inferior position comparison criterion above-mentioned N_BWD frame number and one second arbitrary value N so comprising _bRelatively, N _bRepresented the piece B before current block _N-1Before in succession with the frame number of " oppositely " pattern analysis.

Positive response for the comparison of in testing procedure 1118, carrying out, follow step 1118a in this test back, it comprises to the stationarity parameter value STAT* (n) of middle stationarity parameter value STAT* (n) tax with the piece before current block, deduct a determined value again, this determined value depends on the second arbitrary value N _b, this determined value is designated as f ₂(N_BWD).As can be seen, at assignment step 1118a, as a result of, middle stationarity parameter value reduces.

Yet, if relatively make negative response for the inferior position of carrying out at testing procedure 1118, so step 111 is included in step 1118b the stationarity parameter value of the piece before the current block promptly, stationarity parameter value STAT* (n) in the middle of STAT (n-1) is appointed as.

In Fig. 2 b, can notice that assignment step 1118a and 1118b back be a step and then, the label of this step is 1118c, it is using zero to substitute with the piece number of " oppositely " lpc analysis mode treatment in succession, and the step of this zero setting can be upgraded the whole middle steadily process of parameter value of calculating.

If for the response of compare test step 1112 negate, the transition of " forward direction " lpc analysis does not then take place, the previous piece B of handle in step 1119 _N-1Stationarity parameter value STAT (n-1) compose to give middle stationarity parameter value STAT* (n).

In the ending of step 111, for current block B _nStationarity parameter value STAT* (n) in the middle of setting.

With regard to comprising the step 112 of regulating above-mentioned middle stationarity parameter value, by noting referring to Fig. 2 b, it preferably includes step 1120, and to differentiate the prediction gain of " oppositely " LPC filtering and " forward direction " LPC filtering, these yield values are designated as Gpb and Gpf respectively.Should be understood that above-mentioned discriminating includes only respectively for above-mentioned " forward direction " and " oppositely " filtering storage and reads yield value.Except above-mentioned yield value, step 1120 can comprise difference or the ratio of the fiducial value (being designated as DGfb) of prediction gain for above-mentioned " forward direction " and " oppositely " prediction gain.

As shown in still Fig. 2 b, the step 112 of Fig. 2 a has comprised the step 1121 after above-mentioned steps 1120, this step comprise with improvement value Δ S change in the middle of stationarity parameter value STAT* (n), according to the noticeable especially feature of method of the present invention, this improvement value is the function of the fiducial value of " forward direction " and " oppositely " LPC filter forecasting gain.

Generally, can point out that the function of representing improvement value Δ S can be designated as

ΔS＝f _r(Gpf，Gpb)

Here, as before, Gpf and Gpb point out the gain of " forward direction " and " oppositely " LPC filter forecasting respectively.

Generally, can point out,, can set up the function f of improvement value Δ S according to considering the fiducial value direction _r(Gpf Gpb) is a function that increases and decreases with this fiducial value respectively.When " oppositely " LPC filter gain value that ratio is pointed out to compare with " forward direction " LPC filter gain, this selection can at random keep and not destroy the general aspects of the method (purpose of the present invention), so function f _rBe increased to above-mentioned fiducial value DGfb.Under opposite situation, function f _rReduce.

In other words, by increasing or reduce the middle stationarity parameter value of improvement value Δ S, change is proportional with this fiducial value of gain.Generally, this change is write as STAT (n)=STAT* (n)+k Δ S.In fact, k gets and does 1.Say more more specifically, its expression, when the gap between " forward direction " and " oppositely " LPC prediction gain increased, the algebraic value of improvement value Δ S increased, so function f _r(Gpf is an increasing function Gpb), and when above-mentioned same gap reduces, and the algebraic value of improvement value Δ S reduces, and above-mentioned gap is determined to be between the prediction gain of the prediction gain of LPC " oppositely " filtering and LPC " forward direction " filtering.In fact, the determining function according to the definition in this gap is increasing function or decreasing function.

Therefore, shown in Fig. 2 b, in the ending of step 1121, so for k=1, stationarity parameter value STAT* (n) in the middle of can proofreading and correct with the algebraic value of above-mentioned improvement value Δ S is to calculate stationarity parameter value STAT (n).

After step 1121, in step 1122, set stationarity parameter value STAT (n).

Now 2c provides the more detailed description of the step 1121 of Fig. 2 b with mode of priority in conjunction with the accompanying drawings, wherein, considers the computation process of stationarity parameter, is applied to the improvement value and is applied to LPC " forward direction " and the as many test criteria of " oppositely " prediction gain value.

As shown in above-mentioned Fig. 2 c, step 1121 can comprise first step 1121a, and this step is according to the function f of mentioning in the past _r(Gpf, Gpb) calculating of startup improvement value Δ S.In instructions, will provide the example of not taking charge of of available functions below.

At first, in step 1121b, improvement value Δ S will make odds ratio with value 0, and in fact, this compare test can be determined the increase of improvement value Δ S.

If make positive response for above-mentioned testing procedure 1121b, then improvement value Δ S is for just, and it is corresponding to the increase of the fiducial value of " forward direction " and " oppositely " LPC filter forecasting gain, stand the superiority condition check of the yield value of " oppositely " LPC filtering again according to the step of stationarity parameter value in the middle of the improvement value Δ S increase, should value and one first on the occasion of making comparisons, described first on the occasion of being to determine in the comparative of superiority that " oppositely " LPC filter gain value Gpb and this determined first made comparisons on the occasion of (being called Si).

If make negative response, then in step 1121g, middle stationarity parameter STAT* (n) is composed value STAT (n) to stationarity parameter for above-mentioned testing procedure 1121c.

If make positive response for above-mentioned testing procedure 1121c, the inferior position conditional test of stationarity parameter value STAT* (n) in the middle of the increase of the middle stationarity parameter value of then improvement value Δ S stands again, it and one second are determined to make comparisons stationary value of STATi natural representative on the occasion of STATi.This inferior position test condition is carried out in step 1121e.

If make negative response, then in above-mentioned steps 1121g, middle stationarity parameter value STAT* (n) is composed to middle stationarity parameter value STAT (n) for above-mentioned testing procedure 1121e.

If make positive response for inferior position test condition 1121e, then at step 1121i, but increased positive value delta S in stationarity parameter STAT* (n) compose to middle stationarity parameter value STAT (n).

On the contrary, if make negative response for above-mentioned testing procedure 1121b, then improvement value Δ S bears, the step that middle stationarity parameter reduces improvement value Δ S (this value is born) stands the check of inferior position test condition again in comparison step 1121d, wherein, " oppositely " LPC filter gain value Gpb and one the 3rd making comparisons of determining on the occasion of (being called Sd).This 3rd determine on the occasion of the natural representative LPC filter gain value.

If make negative response, then middle stationarity parameter STAT* (n) is composed to stationarity parameter value STAT (n) at step 1121g for above-mentioned testing procedure 1121d.

On the contrary, if above-mentioned testing procedure 1121d is made positive response, the minimizing step of stationarity parameter minimizing improvement value Δ S also stands the superiority condition check then in being called the compare test step of 1121f, wherein, the value of intermediate stable parameter S TAT* (n) and one the 4th making comparisons of determining on the occasion of (calling STATd).Certainly, the 4th determine on the occasion of having represented the stationarity parameter value of selecting.

If make negative response, then middle stationarity parameter STAT* (n) is composed to stationarity parameter STAT (n) at step 1121g for above-mentioned testing procedure 1121f.

On the contrary, if provide positive response for above-mentioned testing procedure 1121d, then in being called the compare test step of 1121f, the step that middle stationarity parameter value reduces improvement value Δ S also can stand the superiority condition check, wherein, middle stationarity parameter value STAT* (n) with one be called that the 4th of STATd determines on the occasion of making comparisons.Certainly, the 4th determine on the occasion of having represented the stationarity parameter value of selecting.

If make not response for above-mentioned testing procedure 1121f, then middle stationarity parameter value STAT* (n) composed to stationarity parameter STAT (n) at step 1121g.

If make positive response for above-mentioned testing procedure 1121f, then the middle stationarity parameter value STAT* (n) of the algebraic value (it is a negative value) that has increased improvement value Δ S is composed and give stationarity parameter STAT (n), so middle stationarity parameter value reduces to set up stationarity parameter value STAT (n) at step 1121h

In ending place of step 1121g, 1121h and 1121i, so in the step 1122 of Fig. 2 b, set stationarity parameter STAT (n).

As for function f _r(Gpf, Gpb), can prove, it can comprise the nonlinear function of the fiducial value of " forward direction " and " oppositely " LPC filter gain, wherein, the fiducial value of " forward direction " and " oppositely " LPC filter forecasting gain itself can be the ratio or the difference of the gain of " forward direction " and " oppositely " LPC filter forecasting.Can use the function (such as linear function) of other types.

At Fig. 2 d nonlinear function f is shown _r(Gpf, first example Gpb).

In the pattern example of Fig. 2 d, along the paired value of " oppositely " LPC filter forecasting gain G pb of ordinate and " forward direction " LPC filter gain Gpf for respectively corresponding to the ratio ρ=Gpb/Gpf of the slope bigger or more less than the slope of straight line Δ S=0, stipulate positive improvement value Δ S (Δ S＞0) or Δ S＜0 of bearing.

In Fig. 2 e, show such situation, wherein, the relative value of " forward direction " and " oppositely " filter forecasting gain is no longer corresponding to gain ratio ρ, but corresponding to the difference of above-mentioned gain.In this case, the relative value of " forward direction " and " oppositely " LPC filter forecasting gain also can be a nonlinear function, this function can be to corresponding to Gpb, the difference regulation improvement value Δ S of the paired value of Gpf, Gpb, the paired value of Gpf is corresponding to some straight lines, the algebraic value of their horizontal ordinate initial point respectively less than in or greater than the initial point of the horizontal ordinate of straight line Δ S=0.In the situation of Fig. 2 e, be parallel to each other as the straight line of the function zoning of the symbol of improvement value Δ S.

According to another special aspects of process (it is a purpose of the present invention), further suggestion (is for example worked as sound signal and is comprised noiseless speech signal) during silent frame, allows not self-adaptation current block B _nThe stationarity index.In this case, be shown in before the step 1111 of step 111 of Fig. 2 b 1111a in steps, this step comprises, for each continuous current block, determine the average energy of audio digital signals, and in this same step,, this average energy and the threshold value of determining of representing silent frame are made comparisons according to the inferior position comparison criterion.In Fig. 2 b, this threshold value is called ENER.SIL.If make positive response, then in the given step that is shown in Fig. 2 b, the steadiness parameter value STAT (n-1) of previous piece is composed the steadiness parameter value STAT (n) that gives current block for above-mentioned testing procedure.In above-mentioned figure, step 1111a and 1111b are shown in broken lines, because they are for for example the speech signal coding being prepared.

Provide the more detailed description that realizes decision function Dn in conjunction with Fig. 2 f now, can obtain decision content dn (n) from this function.Provide description with a kind of preferable form, yet, according to purpose of the present invention, analyze dn (n) value in order to draw continuous selection, this decision function is adaptive in short-term, and this decision function can be compared with the decision function of description in the previously mentioned paper (being delivered by S.PROUST, C.LAMBLIN and D.MASSALOU).

From step 120, for current block B _n, LPC wave filter that at first calculates at current block and previous piece B _N-1Wave filter between distance (be called d _LPC).For example, this distance calculation is carried out with the LSP frequency parameter, as mentioning in the description relevant with the process of describing in the above-mentioned paper.

Be noted that

The value of the threshold value S_PRED that runs in-the criterion (n) s_TRANS, s_sTAT and G1 is to judge according to the prediction gain of " oppositely " and " forward direction " LPC filtering;

Threshold value S_LSP_L that runs in-the criterion and S_LSP_H judge according to the distance between the LSP frequency vector, described LSP frequency vector representative and two continuous piece B _N-1And B _nTwo " forward direction " LPC wave filters relatively;

The prediction gain Gpf of-" forward direction " LPC wave filter;

-" oppositely " the LPC wave filter prediction gain Gpb; And

-according to the prediction gain Gpi of " forward direction " wave filter of the method interpolation that illustrates in the paper of mentioning in the superincumbent instructions.

Referring to Fig. 2 f, set up the criterion that is used to set up decision function with following manner:

If-in succession LPC wave filter is very steady (that is, for d _LPC＜S_LPC_L), if then it is in " oppositely " LPC filter patterns is not just carried out the switching that is filtering to " forward direction " LPC filtering from " oppositely " LPC, its condition is that the prediction gain of " oppositely " LPC wave filter is greater than the prediction gain that deducts " forward direction " LPC wave filter after the s_sTAT value.Point out, select s_sTAT, thereby using by distance d _LPCThe frequency spectrum that records helps the selection of " oppositely " LPC wave filter when very steady;

If-in succession LPC wave filter has significant transition, that is, and for d _LPCIf＞S_LSP_H and Gpf＞Gpb_s_TRANS, so the filter patterns of selecting is " forward direction " LPC filtering, that is, and d _n(n)=0 its value of symbol is " fwd "), otherwise, d _n(n) no better than 1, (its value of symbol is " bwd ").Be noted that and select s_TRANS, thereby using apart from d _LPCMeasure when frequency spectrum transition occurring, help very much the selection of " forward direction " LPC wave filter;

-otherwise, in every other situation, if Gpb＞Gpf_S_PRED and Gpi＞Gpf_S_PRED, then the LPC wave filter of Bao Chiing is through interpolation " oppositely " LPC wave filter, its condition is that the gain of the latter's gain and pure " oppositely " LPC wave filter surpasses threshold value Gi above-mentioned.If the condition of above-mentioned relevant prediction gain value does not satisfy, then select " forward direction " LPC filtering.

For the number of " forward direction " LPC wave filter of increasing transmission, thereby increased the intensity of coded system for transmission error, one as the energy signal E that will encode _n(that is relevant block B, _nEnergy) can advantageously select " forward direction " LPC filter patterns during less than the energy value ENER_SIL of silent frame, the energy value of silent frame is corresponding to minimum audible value.

Described a set condition in Fig. 2 f, they can set up decision function Dn, and draw corresponding selection assay value dn (n), here, are adaptive in short-term to decision function Dn.

For example, can be positioned at 0 place on the yardstick, and corresponding to STAT (n) value very stably corresponding to the STATn value stationarity parameter value STAT (n) of non-stationary, to the yardstick 100.

According to stationarity parameter value STAT (n),, change decision function Dn by self-adaptation for threshold value.

The stationarity of signal increases more, has more to utilize to take " oppositely " LPC filter patterns: S_PRED, S_LSP and S_LSP_H to increase.

As an example that does not limit, by each current LPC piece B for above-mentioned threshold value _nThe change function find out:

S_PRED9n)=f _{S_PRED}(STAT (n)) wherein, function f _{S_PRED}Value with STAT (n) rises;

S_LPC_L (n)=f _{S_LPC_L}(STAT (n)), wherein, function f _{S_LPC-L}Rise;

S_LPC_L (n)=f _{S_LSP_R}(STAT (n)), wherein, function f _{S_LPC-R}Rise.

In the adaptation to above-mentioned threshold value, as can be seen, the increasing function of mentioning for example is such function, and they relate to function f _{S_LPC-L}And f _{S_LPC_H}Function f _{S_PRED}Be the improvement function of variable stationarity parameter, it has following form:

S_PRED(n)＝α·STAT(n)+β

Here α and β are the real numbers between two 0 and 1, and the value of S_PRED (n) is limited in interval [S_PRED _m), S_PRED _M] between, the value that on behalf of two experiments, S_PREDm and S_PREDm determine.When stationarity parameter STAT (n) less than a given threshold value S _FWDThe time, for the risk of limiting filtering device again, so need can select " forward direction " LPC filter patterns.

On the other hand, s_TRANS, s_TAT and G ₁Threshold value remains fixed value, and for example these values can be one-1dB, 5dB and 0dB.

In Fig. 2 f, describe the foundation of decision function Dn in the following manner and analyze selective value d _n(n) acquisition: after above-mentioned steps 120, carry out relevant current LPC piece B _nThe testing procedure 121 of energy, its way is that this energy and noiseless energy value ENER_SIL are made inferior position relatively, perhaps with stationarity parameter value STAT (n) and the S that mentions in instructions _FWDValue is made inferior position relatively.If above-mentioned testing procedure 121 is made positive response, then in step 122, selection assay value dn (n) is taken as 0 (that is value of symbol " fwd ").

If make negative response for above-mentioned testing procedure 121, then carry out a new testing procedure, it is about selecting assay value d _N-1(n-1) with logical value 1 (that is value of symbol " bwd ").

If make positive response for above-mentioned testing procedure 123, then in step 124 pair above-mentioned LPC filtering apart from d _LPCCarry out new test, make this distance and threshold value S_LSP_H (n) make comparisons (making odds ratio) with this threshold value.

If make positive response for above-mentioned testing procedure 124, then realize new testing procedure 126a, it comprises that " forward direction " LPC filter forecasting gain G pf is deducted threshold value s_TRANS with " oppositely " LPC filter forecasting gain G pb to make comparisons afterwards.

If make positive response, then logical value 0 (value of symbol " fwd ") is composed to selecting assay value d for above-mentioned testing procedure 126a _n(n), if make negative response, then value 1 (value of symbol " bwd ") composed and give same selection assay value for above-mentioned testing procedure 126a.Corresponding step is called 128 and 129.

If make negative response, then carry out new testing procedure 125 for above-mentioned testing procedure 124.Testing procedure 125 comprises to be carried out LPC filtering apart from d _LPCMake inferior position relatively with threshold value S_LPC_L (n).

If make positive response for testing procedure 125, then carry out new testing procedure 126b, this testing procedure is to make odds ratio after " oppositely " LPC filter forecasting gain deducts above-mentioned value s_sTAT with the gain of " forward direction " LPC filter forecasting.

If testing procedure 126b is made positive response, then logical value 1 (that is value of symbol " bwd ") is composed to selecting assay value d in step 129 _n(n).

If testing procedure 126b is made negative response, then logical value 0 (that is value of symbol " fwd ") is composed to selecting assay value d in step 128 _n(n).

On the contrary, if make negative response for testing procedure 125, then carry out new test in step 127, this testing procedure comprises: the comparison condition between " forward direction " LPC filter forecasting gain after verifying " oppositely " LPC filter gain Gpb and deducting threshold value S_PRED (n); Middle LPC filter forecasting gain G pi and deduct above-mentioned threshold value S_PRED (n) after " forward direction " LPC filter forecasting yield value between odds ratio; And " oppositely " filter forecasting gain G pb and threshold value G ₁Odds ratio and middle filter forecasting gain G pi and threshold value G ₁Comparison.

Be noted that the negative response that the testing procedure 123 mentioned for the front is made also causes carrying out above-mentioned testing procedure 127 in instructions.

If make positive response, then logical value 1 (that is value of symbol " bwd ") is composed to selecting assay value d in step 129 for above-mentioned testing procedure 127 _n(n), if make negative response, then on the contrary logical value 0 (that is value of symbol " fwd ") is composed to selecting assay value d in step 128 for above-mentioned testing procedure _n(n).

So,, set selection assay value d by using decision function Dn _n(n), it obtains with above-mentioned logical value 1 or 0, and these logical values with as the priority value that is back to " oppositely " or " forward direction " filter patterns of the function of stationarity parameter value or do not have priority value and interrelate.

Provide more detailed description according to the audio digital signals code device of purpose of the present invention in conjunction with Fig. 3 now, its way is in the coded signal that sends, and makes double analysis according to the criterion of " forward direction " and " oppositely " LPC selection analysis respectively.

Especially, be noted that the digital signal that will encode is divided into frame that they are made of the continuous piece of sample, for example, each piece comprises a given number N sample.

In Fig. 3, the mode of the audio digital signals that will encode with continuous sample block Bn structure be not shown, because this mode of operation is known in state-of-the art, and can realize to form simple storage buffer, for example, to its addressing, to read frame rate and piece frequency periodically.

As also as shown in above-mentioned Fig. 3, code device (it is a purpose of the present invention) comprises " forward direction " lpc analysis wave filter (label is 1A) and " oppositely " analysis filter (label is 1B), with the coded signal of transmissibility through transmission, it comprises the LPC filtering parameter, and analyzes and judge indication and Pr parameter (relevant with pumping signal CELP with frequency analysis).

Generally, as can be seen, analyze and judge indication and select assay value d _n(n) corresponding, as instructions above-mentioned.As for the LPC filtering parameter, can point out that (it is the use-pattern of purpose of the present invention, and these parameters are corresponding to specific parameter, and this will be described later according to coding method.

In Fig. 3, also illustrate, in according to code device of the present invention, have one as a function of stationarity parameter value and the sef-adapting filter of working, its label is 1E.Certainly should be understood that sef-adapting filter 1E receives initial digital signal (being called Sn (t)) (that is current block Bn).Wave filter 1E uses filter LPC parameter, and to calculate residual signal, this residual signal is encoded by module 1F again.These LPC parameters and filtering judge that indication has constituted the part of the coded signal that is sent to code translator.

In addition, as shown in Figure 3, code device (it is a purpose of the present invention) comprises the code device (label is 1F) of the residue coded signal that does not send, by res _n(t) Biao Shi residue coded signal directly obtains at the output of sef-adapting filter 1E, then this signal and digital signal is delivered to the coding module place input end of the residue coded signal that does not send, to produce comprehensive residual signal res_syn _n(t).

Liftering module (label is 1G) receives comprehensive residual signal, and transmits integrated signal s_syn _n(t).

Memory module 1H receives above-mentioned integrated signal s_syn _n(t), to current block Bn, the integrated signal that so obtains is expressed as s_syn with the above-mentioned integrated signal of sending previous piece _N-1(t).The label that this integrated signal is delivered among above-mentioned Fig. 3 is " oppositely " lpc analysis wave filter of 1B.

As shown in Figure 3, code device (it is a purpose of the present invention) can be carried out the coding of audio digital signals, that is, and and for the non-stationary district, to encoding from the above-mentioned audio digital signals of " forward direction " LPC wave filter; And for plateau region, then to the above-mentioned integrated signal s_syn from " oppositely " LPC wave filter 1B _N-1=(t) encode, as described below.

As will in above-mentioned Fig. 3, seeing, for this purpose, code device (it is a purpose of the present invention) comprises computing module IC for each current LPC piece Bn, it is according to the stationarity degree of stationarity parameter calculating audio digital signals, and the value of described stationarity parameter is between maximum stationarity value and minimum stationarity value.Certainly, according to cataloged procedure (it is a purpose of the present invention), stationarity parameter is exactly the parameter S TAT (n) that describes at instructions in front.The front has also defined maximum stationarity value and minimum stationarity value.

As also in Fig. 3, going out to illustrate, code device (it is a purpose of the present invention) comprises a module that is called ID, be used for setting up judgement letter LPC and select assay value according to aforementioned stable parameter S TAT (n), mention in instructions as the front, decision function is called Dn and LPC selects assay value certainly corresponding to the d that is called that has mentioned in instructions _n(n) LPC selects assay value.Can remember, select assay value d _n(n) fixing value 0 or 1 (logical value), they analyze value of symbol " fwd " and " bwd " corresponding to the selection that is used for " forward direction " and " oppositely " lpc analysis respectively.

Especially, should be understood that the foundation that relates to decision function Dn is to realize that with for example software (Fig. 2 f was described such as before combining) is corresponding.In addition, all as shown in Figure 3 comprise that according to code device of the present invention LPC filter analyses identification module (is called 1D ₂), this module receives selects assay value d _n(n), and for current LPC piece Bn, can send respectively LPC " oppositely " and " forward direction " filtering parameter as the function of above-mentioned assay value.Should understand clearly that " oppositely " LPC filter analyses and " forward direction " LPC filter analyses parameter wave filter that certainly label is respectively 1B and 1A in Fig. 3 is sentenced digital form and obtained.Respectively these parameters being appointed as about label is the Af of " forward direction " lpc analysis wave filter " forward direction " LPC filter analyses parameter of 1A _n(z) and about label the Ab of " oppositely " lpc analysis parameter of " oppositely " lpc analysis wave filter of 1B _n(z).Respectively these parameters are delivered to module 1D ₁And 1D ₂

As for identification module 1D ₁The generation of equipment for example can show that it comprises two different storage areas with a kind of infinite mode, and they can store filtering parameter Af respectively _n(z) and Ab _r(z), as selecting assay value d _nThe selection assay value d of the function of current logical value (n) (0 or 1) _n(n) for example module 1D that makes it possible to serve as reasons ₂Read the filtering parameter value of storage and addressing, and transmit these filtering parameters by the latter.

At last, as shown in Figure 3, as can be seen, in order to be the sef-adapting filter of IE according to stationarity Value Operations label, can realize transport function A (z) by filter element according to the code device of purpose of the present invention, this transport function is that basis is by above-mentioned identification module 1D ₂The filtering parameter value of sending is set up.

Will understand that also auto adapted filtering module 1E can reach with the wave filter with adjustable coefficient, the coefficient value of this wave filter is by above-mentioned identification module 1D ₂Send.So the filtering of being undertaken by module 1E is self-adaptation type, its is worked as the function of the stationarity degree of the audio digital signals that will encode.So module 1E is according to initial audio digital signals s _n(f) send LPC filtering residual signal and (use res _n(t) expression) to residual signal coding module 1F, then, this module can be sent and be used res_syn _n(t) the comprehensive residual signal of Biao Shi LPC.

At last, module 1G is a filtration module, and its transport function is the inverse of the transport function of module 1E, and it draws from the stored parameters of module 1E.It receives the comprehensive residual signal res_syn of LPC that is sent by the coding module of coding residual signal _n(t), here, the coding residual signal is sent by module 1F.So can understand audio digital signals s _n(t) coding is analyzed by LPC " forward direction " and " oppositely " respectively in module 1E and is carried out, and LPC " forward direction " and " oppositely " are analyzed by LPC " forward direction " and " oppositely " analysis filter 1A and 1B implementation, when selecting assay value d _nWhen (n) having the indication (that is, selecting the indication of the value of before having quoted of analysis) of value of symbol " fwd " and selection analysis, coded signal s_c _n(t) be present in the transmission of " forward direction " LPC filtering parameter.This mode of operation makes it possible to carry out the coding of audio digital signals, and function as the stationarity degree of digital signal, help respectively it to be remained on one of " forward direction " and " oppositely " LPC filter patterns, and further limit switching number from a kind of filter patterns of considering to another kind of pattern.

The code translator of the coded audio digital signal that adopts double analysis is described in conjunction with Fig. 4 now.Described double analysis is according to the criterion of " forward direction " and " oppositely " lpc analysis the coded signal that transmits to be carried out respectively according to the coding method as purpose of the present invention.And coding is to use to carry out such as the code device among Fig. 3.

Generally, as can be seen, for each lpc analysis piece, the coded signal s_c of transmission _n(t) comprise above-mentioned selection assay value, and under such situation, wherein for the lpc analysis piece of being considered, select assay value corresponding to " forward direction " lpc analysis, then coded signal comprises the coding parameter (Pr of " forward direction " LPC filtering parameter and LPC filtering residual signal _nParameter), that is, residue coding module 1F is at comprehensive residual signal res_syn _n(t) the signal res in _n(t) divide coding parameter.

As shown in Figure 4, as can be seen, code translator comprises the comprehensive module (label is 2A) of filtering residual signal at least, and it receives the coding parameter of the LPC residual signal of being sent by module 1F.Module 2A is to the decoding of the coding parameter that provided by module 1F, and thereby send comprehensive residual signal, in Fig. 4, it is designated as res_sy _n(t).

Code translator as shown in Figure 4 comprises that also a label is the contrary auto adapted filtering module (as the function of stationarity degree) of 2B, and its receives the comprehensive residual signal of being sent by module 2A before quoted, and can produce integrated signal s_syn _n(t) (it represents audio digital signals), and this integrated signal in fact constitutes decoded signal.

Certainly should be understood that, when the transmission filtering parameter, because this fact of transmission, the filtering parameter that is received by code translator that liftering module 2B uses is " forward direction " lpc analysis parameter, it is also to be understood that, analyzes and judges corresponding to " forward direction " lpc analysis, perhaps, on the contrary, corresponding to " oppositely " filter analyses, as will be described below.

For this reason, code translator (it is a purpose of the present invention) comprises that certainly label is 2D " oppositely " LPC filtration module, and it is received in the integrated signal of the LPC piece before the current LPC piece, and the integrated signal of current block is designated as s_syn _nAnd in Fig. 4, be designated as s_syn (t), at the integrated signal of the piece of its front _N-1(t).For this reason, should be understood that about current block B _nAnd be designated as s_syn _n(t) so integrated signal is delivered to " oppositely " filtration module 2D by memory module (label is 2E), in fact by for the self-adaptation addressing of reading, reading of integrated signal can be moved to the integrated signal of corresponding last (current block Bn).

At last, for guaranteeing above-mentioned mode of operation, as shown in Figure 4, code translator (it is a purpose of the present invention) also comprises the identification module with label 2C, and this module can be carried out " forward direction " and " oppositely " LPC discriminatory analysis respectively.On the one hand, for control is differentiated, module 2C receives received selection assay value (that is value d, _n(n))), on the other hand, receive " forward direction " LPC filtering parameter (that is parameter A f that, is sent out _n(z)), and by module 2D obtain " oppositely " LPC filtering parameter Ab _n(z).Then, as the function of selecting assay value (that is value d, _n(n)), module 2C can send " forward direction " filtering parameter Af as the function of stationarity degree _n(z) or " oppositely " filtering parameter Ab _n(z).To contrary auto adapted filtering module 2B.

As for the specific embodiment of module 2C and 2B, can mention that these modules can only comprise roughly the module 1D with Fig. 3 ₂Identical with 1E, perhaps, more particularly, the module roughly the same with the mould 1G of Fig. 3.

As for according to purpose of the present invention the effective embodiment (it can use the process of describing in front the instructions) of code device, made two kinds of concrete forms.

^* According to telephone band CELP type scrambler with the height output expansion of the UIT-T standard of 8kh/s

The actual coding device is made of CELP type telephone band (from 300 to 3400Hz) scrambler, be output as 12kb/s according to the technology of before in instructions, mentioning that is called ACELP, for the excitation that provides by algebraically codepages, configuration frame in the duration of 10ms.

Per 80 samples, " forward direction " lpc analysis are that rank are 10 analysis, are 30 analysis and " oppositely " lpc analysis is rank.

The sub-piece that coding of residual signal has been divided into 40 samples.Each piece Bn comprises 80 samples.

The self-adaptation of stationarity parameter STAT (n)

Above-mentioned stationarity parameter value is at two extremums 0 and 100 (that is above-mentioned value STAT, _mAnd STAT _M) between change.

The previous auto-adaptive function of describing at instructions, particularly function f _a(N_BWD) and f _b(N_BWD) be:

In these relational expressions, x=Dgfb.

As for the function f r that can set up the improvement value Δ S that before mentions in instructions, it is the step function (step fuction) of variable x, x=Gpb-Gpf wherein, and Δ S=f _r(x) and have a following value:

A nearly step of the adjusting of STAT (n) suffers following condition (they are mentioned in conjunction with Fig. 2 C in front): if Δ S＞0:

If STAT* (n)＜STATi, then STAT (n)=STAT* (n)+Δ S

Other situations, then STAT (n)=other situations of STAT* (n):

STAT (n)=STAT* (n) is STATi=40.6 here.

In this form, do not use other test conditions that are marked with 1121d, 1121c and 1121f among Fig. 2 c. The self-adaptation of decision threshold

As for decision threshold:

Adopt S_PRED in the following manner:

S_PRED(n)＝0.03STAT(n)+1.0

S_PREDE[S_PRED _m, S_PRED _M], SPRED _m=1.63 and SPRED _M=4;

Make threshold value S_LSP_L self-adaptation with following step function:

If measure the LPC stability of filter with threshold value S_LSP_L, then threshold value s_sTAT be fixed on 4.0dB.

In this form without threshold value S_LSP_H.

G ₁Threshold value be fixed on 0dB.

As for the energy value ENER-SIL that characterizes with silent frame, this value is fixed on 40dB, measurement in the scope of 80 sample S (i) of current block Bn here: $\mathrm{ENER}\_\mathrm{SIL}=10\mathrm{Log}\left[\underset{i=0}{\overset{i<80}{\mathrm{\&Sigma;}}}S{\left(i\right)}^2\right]$

Previously mentioned S _FWDThe value of threshold value is set in 40.6, is used for when STAT (n) value is lower than this threshold value, and still further restricted passage applies the risk that " forward direction " LPC filter patterns switches.

^* Carry out CELP widening frequency band scrambler with two sub-frequency bands 16/24/32kg/s with following condition Second kind of form:

The widening frequency band scrambler of-0 to 7000Hz (being divided into two sub-frequency bands).The dominant frequency band is with there being the CELP technology for encoding, and frame has 120 samples, produces excitation by algebraically codepages, and is transmitted in master (host) some energy of frequency band and spectral characteristic between 6000Hz and the 7000Hz.

-per 120 samples are made " forward direction " lpc analysis or are made " oppositely " lpc analysis with 50 coefficients with 14 coefficients.In " forward direction " lpc analysis, be divided into the LPC piece of two 60 samples, the wave filter that is used for the 1st sub-piece is made interpolation according to working as pre-filter and first pre-filter.

The calculating of stationarity parameter STAT (n)

In this form, above-mentioned stationarity parameter is at two extremums 0 and 120 (that is above-mentioned STAT, _mAnd STAT _MValue) changes between.

As for stationarity parameter value self-adaptation, fa (N_BWD) and f _b(N_BWD) value of function is:

As for the f that allows to be set in the improvement value AS that mentions in the instructions of front _rFunction, it is the step function of a variable x, x=Gpb/Gpf here, and Δ S=f _r(x) and have a following value:

The adjusting of STAT (n) further stands following conditional test (they are mentioned in conjunction with Fig. 2 c in front): if Δ S＞0:

If Gpb＞S:

If STAT* (n)＜STATi, then STAT (n)=STAT* (n)+Δ S

If other situations, then STAT (n)=STAT* (n)

If other situations, then STAT (n)=STAT* (n) is if other situations:

If STAT* (n)＜STATi, then STAT (n)=STAT* (n)+Δ S

If other situations, STAT (n)=STAT* (n) STATi=80 wherein then, Si=0dB.In this form, not be used in other test conditions that indicate 1121h and 1121d among the figure c. The self-adaptation of decision thresholdAs for decision threshold: make S_PREDm self-adaptation S_PRED (n)=0.03 STAT (n)-0.5 in the following manner, be confined to [S_PRED at interval _m, S_PRED _M] in, S_PRED wherein _m=0.5, and S_PRED _M=2.5.Make the S_LSP_L threshold adaptive by means of following step function:

Make the S_LSP_H self-adaptation with following step function:

In the situation by means of the LPC wave filter transition of S_LSP_H threshold measurement, used s_TRNS threshold setting is at 0dB.

In the situation by means of the stationarity of the LPC wave filter of S_LSP_L threshold measurement, used s_sTAT threshold setting is at 2.5dB.

The threshold setting of G is at 0dB.

As for the energy value ENER_SIL that characterizes with silent frame, this value is fixed on 50dB, and this is to measure in the scope of 120 sample S (i) of current block: $\mathrm{ENER}\_\mathrm{SIL}=\mathrm{Log}\left[\underset{i=0}{\overset{i<120}{\mathrm{\&Sigma;}}}S{\left(i\right)}^2\right]$

Previously mentioned S _FWDThreshold setting is used for when STAT (n) value is lower than this threshold value 60, and still further restricted passage applies the risk that " forward direction " LPC filter patterns switches.

Claims (14)

1. method to digital audio signal coding, the selection criterion that described method is analyzed according to LPC " forward direction " and " oppositely " respectively, adopt double analysis, described audio-frequency signal coding is the coded signal of transmission and the residue coded signal that does not transmit, the coded signal of described transmission comprises the LPC filtering parameter and analyzes determination information, described digital audio and video signals is divided into frame, promptly, the continuous blocks of the sample of defined amount, by " forward direction " LPC filtering described digital audio and video signals is carried out coding for the non-stationary zone respectively, encode with by " oppositely " LPC filtering the integrated signal that obtains from described residue coded signal being carried out for plateau region, it is characterized in that, comprise for the described selection criterion of each current block of a series of described current blocks that constitute present frame:

-determine the stationarity degree of described digital audio and video signals according to stationarity parameter, the value of described stationarity parameter is between maximum stationarity value and minimum stationarity value;

-according to described stationarity parameter, set up by the analysis selective value of judging that letter draws;

-apply described analysis selective value to " forward direction " LPC filtering, thereby encode to described integrated signal with described " oppositely " LPC filtering to described digital audio signal coding and for plateau region with described " forward direction " LPC filtering in the non-stationary zone for described sound signal, make to help described digital audio and video signals filtering to be remained in one of described " forward " and " oppositely " filter patterns respectively, and limit the number of described filter patterns from a kind of to alternative switching or reverse switching according to relation with the stationarity degree.

2. the method for claim 1, it is characterized in that, described decision function is an auto-adaptive function, according to described stationarity parameter each described current block is carried out judgement, the function that the described implementation of described auto-adaptive function can be used as the stationarity degree of described digital audio and video signals helps respectively described digital audio and video signals filtering to be remained in one of described " forward direction " and " oppositely " filter patterns, limits the number of described filter patterns from a kind of to alternative switching or reverse switching thus.

3. the method for claim 1 is characterized in that, the described analysis selective value of setting up according to described decision function is respectively corresponding to " forward direction " LPC filter patterns priority value and " oppositely " LPC filter patterns priority value.

4. the method for claim 1 is characterized in that, the step of stationarity degree that begins to constitute each described current block of the described digital audio and video signals of regulation from the initial value arbitrarily of described stationarity parameter comprises:

-for described current block, as the described analysis selective value of defined amount and the function of the stationarity parameter value before described current block, stationarity parameter value in the middle of calculating, those described analysis selective values are to draw for the different continuous blocks before the described a series of described current block; And

The function of the prediction gain value of " forward direction " and " oppositely " LPC filtering of the frame of-conduct before described present frame is regulated described middle stationarity parameter value.

5. method as claimed in claim 4 is characterized in that, for each described current block, constitutes the step of calculating described middle stationarity parameter value and comprises:

-to the piece before described current block, between described " forward direction " LPC or " oppositely " lpc analysis pattern, make discriminating; And

-for any previous piece with described " oppositely " lpc analysis pattern analysis:

^*Regulation connects the number with the previous frame of described " oppositely " lpc analysis pattern analysis,

^*According to the advantage comparison criterion, the number of described previous frame and first arbitrary value to be made comparisons, described first arbitrary value is represented a plurality of successive frames with described " oppositely " LPC pattern analysis, and if described odds ratio is made positive response,

^*After then the stationarity parameter value at the piece before the described current block being added particular value into the function of described first arbitrary value, compose give described in the middle of the stationarity parameter value, and if described odds ratio is made negative response,

^*Then the stationarity parameter value of the piece before the described current block is composed to stationarity parameter value in the middle of described; And

-for any previous piece with the pattern analysis of described " forward direction " lpc analysis:

^*According to test criteria determine whether the transition from described " oppositely " lpc analysis pattern to " forward direction " lpc analysis pattern takes place between piece before the described previous piece and described previous piece, and if make positive response for described generation whether test,

^*Then according to the inferior position comparison criterion, the number of described previous frame and second arbitrary value are made comparisons, the continuous frame number of usefulness described " oppositely " the LPC pattern analysis of described second arbitrary value representative before described previous piece, and if described inferior position is relatively made positive response

^*Compose to stationarity parameter value in the middle of described after then the stationarity parameter value of the piece before the described current block being deducted a setting, described setting is the function of described second arbitrary value, and if described inferior position is relatively made negative response,

^*Then the stationarity parameter value of described previous piece is composed to stationarity parameter value in the middle of described.

6. method as claimed in claim 4 is characterized in that, the step that constitutes for the described middle stationarity parameter value of each described current block adjusting comprises:

-between the prediction gain of described " forward direction " LPC filtering and " oppositely " LPC filtering, differentiate;

-make described middle stationarity parameter value change an improvement value, described improvement value is the function of described " forward direction " and the relevant prediction gain value of " oppositely " LPC filtering, the variation of described middle stationarity parameter value, promptly, increasing or reduce, is proportional with described improvement value.

7. method as claimed in claim 6, it is characterized in that, with described improvement value increase pro rata described in the middle of the step of stationarity parameter value also stand described " oppositely " LPC filter gain with respect to first regulation on the occasion of the superiority condition check and described intermediate stable parameter value with respect to second stipulate on the occasion of the inferior position conditional test.

8. method as claimed in claim 6, it is characterized in that, with described improvement value reduce pro rata described in the middle of the step of stationarity parameter also be subjected to described " oppositely " LPC filter gain with respect to the 3rd regulation on the occasion of the inferior position conditional test and described in the middle of the stationarity parameter value with respect to the 4th regulation on the occasion of superiority condition check.

9. as any one described method among the claim 6-8, it is characterized in that the relevant value of described " forward direction " and " oppositely " LPC filter forecasting gain comprises ratio or the difference between the prediction gain of described " forward direction " and " oppositely " LPC filtering.

10. the method for claim 1 is characterized in that, also comprises for each described continuous current block:

-set up the average energy of described digital audio and video signals,

-according to the inferior position comparison criterion, described average energy and defined threshold to be made comparisons, this threshold value is represented silent frame, and if described inferior position is relatively made positive response,

-then the stationarity parameter value of described previous piece is composed stationarity parameter to described current block.

11. arbitrary described method as claimed in claim 2, it is characterized in that, for be used between minimum value and the maximal value the described stationarity degree of described stationarity parameter, stationarity degree according to the increase of described digital signal, the described auto-adaptive function that constitutes described decision function is the increasing function of the priority value of described " oppositely " LPC filter patterns, here, described minimum value is represented the stationarity degree of the digital signal of non-stationary substantially, and described maximal value is represented substantially the stationarity degree of signal stably.

12. device to digital audio signal coding, the selection criterion that described signal is analyzed according to " forward direction " and " oppositely ", adopt double analysis to be encoded respectively, the coded signal of described audio-frequency signal coding for transmission, the frame that described digital signal is divided into the continuous blocks formation of the sample that comprises defined amount, described code device comprises " forward direction " lpc analysis wave filter and " oppositely " LPC wave filter and the coding residual signal apparatus for encoding to not transmitting, described " forward direction " and " oppositely " LPC wave filter can be sent the coded signal of transmission, this signal comprises LPC filtering number and analyzes and judge indication, and describedly can produce comprehensive residual signal to coding residue apparatus for encoding, the coding of described digital audio and video signals is respectively to carrying out from the described digital audio and video signals in the non-stationary zone of described " forward direction " wave filter and to the described integrated signal from the plateau region of described " oppositely " LPC wave filter, it is characterized in that, for each described current LPC piece, described code device also comprises:

-calculate the device of the stationarity degree of described digital audio and video signals according to stationarity parameter, the value of described stationarity parameter is between minimum stationarity value and maximum stationarity value;

-according to the device of stationarity parameter setting function, it can set the lpc analysis selective value;

-differentiate the device of lpc analysis, it receives described analysis selective value, and for described LPC current block, can send " oppositely " and " forward direction " LPC filtering parameter respectively as the function of described analysis selective value;

-adaptive filter device, receive described digital audio and video signals and receive described " forward direction " and " oppositely " LPC filtering parameter value respectively as the function of stationarity degree as the function of described analysis selective value, and send the coding residual signal to the code device of described coding residual signal, it can be to described digital audio signal coding, and help relevantly respectively described digital audio and video signals filtering to be remained in one of described " forward direction " and " oppositely " filter patterns, and limit the number of described filter patterns from a kind of to alternative switching or reverse switching with the stationarity degree of described digital signal.

13. code device as claimed in claim 12 is characterized in that, for each described lpc analysis piece, the coded signal of described transmission comprises:

-described assay value,

And for the described lpc analysis piece of being considered, in the situation of described analysis selective value corresponding to described " forward direction " lpc analysis:

-described " forward direction " LPC filtering parameter.

14. device to digital audio and video signals decoding, described signal is according to described " forward direction " and " oppositely " lpc analysis, adopt double analysis to be encoded into the coded signal of transmission respectively, this signal comprises the LPC filtering parameter and analyzes and judge indication, it is characterized in that, for each lpc analysis, the coded signal of described transmission comprise described analysis selective value and for the lpc analysis piece of being considered corresponding to " forward direction " lpc analysis in " forward direction " LPC filtering parameter, described code translator comprises at least:

-being used for integration unit to residual signal filtering, it receives the coding parameter of described LPC residual signal and sends comprehensive residual signal,

-as the liftering self-reacting device of the function of stationarity degree, it receives described comprehensive residual signal and can produce the integrated signal of the described digital audio and video signals of representative and constitute decoded signal,

-" oppositely " the lpc analysis device, it receives described integrated signal and can produce " oppositely " LPC filtering parameter,

The device of-do discriminating between described " forward direction " lpc analysis and " oppositely " lpc analysis, it is on the one hand in order to differentiate that control receives described analysis selective value, and on the other hand, receive described " forward direction " LPC filtering parameter and " oppositely " LPC filtering parameter respectively, and can send as the function of stationarity degree described " forward direction " LPC filtering parameter or " oppositely " LPC filtering parameter described liftering self-reacting device extremely as the function of described analysis selective value.

Images